Prospective cohort links thyroid markers to lung cancer immunotherapy toxicity

In a prospective, single-center cohort of 312 patients with advanced non-small-cell lung cancer receiving immune checkpoint inhibitors, baseline thyroid autoantibody positivity and higher baseline thyroid-stimulating hormone levels were associated with more thyroid-related immune adverse events. During median follow-up of 6.4 months, 87 patients, or 27.9%, developed a thyroid immune-related adverse event. The findings identify baseline thyroid status as a potential marker of treatment-related risk, but the investigators characterized the results as exploratory and called for external validation before clinical use.

The study consecutively enrolled treatment-naive adults with histologically or cytologically confirmed stage IIIB or stage IV non-small-cell lung cancer at Huangshi Central Hospital in China between January 2021 and June 2026. Participants had normal thyroid function and were eligible for programmed death protein-1 or programmed death-ligand 1 blockade, either alone, in combination with another checkpoint inhibitor, or alongside conventional anticancer treatment. The researchers excluded patients with pre-existing thyroid disease, other autoimmune disorders, previous checkpoint-inhibitor treatment, pregnancy or lactation, or immunosuppressive treatment. Of 339 people initially enrolled, 27 were excluded after withdrawal, loss to follow-up, transfer to another hospital or incomplete key data, leaving 312 in the final analysis.

Before the first checkpoint-inhibitor infusion, investigators measured serum thyroid-stimulating hormone, free triiodothyronine, free thyroxine, thyroid peroxidase antibodies and thyroglobulin antibodies using chemiluminescent immunoassays. Thyroid function and antibody testing were repeated every six weeks, with an allowable window of two weeks on either side. A thyroid peroxidase antibody concentration above 40 IU/mL or a thyroglobulin antibody concentration above 95 IU/mL was considered positive according to the manufacturer’s assay specifications. Seroconversion was defined as an antibody exceeding its assay threshold during follow-up in a patient whose corresponding baseline result had been negative.

Thyroid immune-related adverse events were classified from biochemical patterns. Hypothyroidism consisted of elevated thyroid-stimulating hormone with decreased free thyroxine, or thyroid-stimulating hormone of at least 10 mIU/L requiring levothyroxine replacement. Hyperthyroidism was defined by decreased thyroid-stimulating hormone with elevated free triiodothyronine or free thyroxine, while subclinical dysfunction comprised abnormal thyroid-stimulating hormone with normal free hormone levels. An endocrinologist reviewed abnormal results and excluded alternative explanations, including non-thyroidal illness syndrome, medications and other underlying conditions. Among the 87 events, 62 were hypothyroidism, 13 were hyperthyroidism and 12 were subclinical dysfunction. Median time to onset was 12 weeks for hypothyroidism, six weeks for hyperthyroidism and 15 weeks for subclinical abnormalities.

Baseline thyroid peroxidase antibodies were positive in 59 patients, or 18.9%, and thyroglobulin antibodies were positive in 44, or 14.1%. The thyroid adverse-event rate was 57.6% among patients positive for thyroid peroxidase antibodies, compared with 20.9% among those negative for that antibody. For thyroglobulin antibodies, the corresponding rates were 61.4% and 22.4%; both differences had P values below 0.001. Patients positive for both antibodies had an event rate of 76.0%, compared with 19.2% among those negative for both and 43.4% among those positive for only one antibody. The median baseline thyroid-stimulating hormone concentration was 2.57 microinternational units per mL, or mIU/L, and was higher in the antibody-positive group than in the antibody-negative group.

Accounting for death as a competing event, the investigators estimated a higher cumulative incidence of thyroid adverse events among antibody-positive patients throughout follow-up. In a Fine-Gray model, baseline antibody positivity was associated with a subdistribution hazard ratio of 3.353, with a 95% confidence interval of 2.256 to 4.985 and P below 0.001. The proportional-hazards assumption was violated for tumor stage and baseline antibody positivity, prompting the use of a stratified Cox model with tumor stage as the stratification factor and a time-dependent interaction term for antibody status. In the final model, antibody positivity at baseline had a hazard ratio of 17.529, with a 95% confidence interval of 7.339 to 41.866 and P below 0.001. A significant time interaction indicated that this association decreased over follow-up, so the large early estimate should be understood as a model result from this cohort rather than a universal multiplier of risk.

Baseline thyroid-stimulating hormone also carried information about subsequent toxicity. In the multivariable model, higher baseline thyroid-stimulating hormone was independently associated with thyroid immune-related adverse events, with a hazard ratio of 2.471 and a 95% confidence interval of 1.845 to 3.309; the analysis did not specify the unit scaling of this continuous covariate. Receiver operating characteristic analysis produced an area under the curve of 0.818, with a 95% confidence interval of 0.763 to 0.873. An exploratory threshold of 2.985 mIU/L, selected by maximizing the Youden index, had sensitivity of 71.3% and specificity of 81.3%. Its positive predictive value was 59.6%, with a 95% confidence interval of 50.0% to 68.7%, and its negative predictive value was 88.0%, with a 95% confidence interval of 82.9% to 91.6%. When the threshold was included in the Cox model, the dichotomized baseline thyroid-stimulating hormone variable was associated with events at a hazard ratio of 4.304, with a 95% confidence interval of 2.644 to 7.004 and P below 0.001.

The timing of post-treatment antibody changes complicated their interpretation. Among patients initially negative for thyroid peroxidase antibodies, 12.6% seroconverted; among those initially negative for thyroglobulin antibodies, 9.7% did so. The median time to seroconversion was 12 weeks. Of the 234 patients negative for both antibodies at baseline, 36, or 15.4%, later crossed the threshold for at least one antibody. Twenty-two of those 36 also developed a thyroid adverse event, but in nine the adverse event occurred first, in six seroconversion occurred first and in seven both were recorded on the same day. In crude comparisons, seroconversion was associated with a higher event incidence, but that association did not persist after the timing of seroconversion was addressed analytically.

Using seroconversion as a time-dependent covariate in the subgroup of patients negative for both antibodies at baseline, the investigators found no independent association with subsequent thyroid immune-related adverse events. The estimated hazard ratio was 0.236, with a 95% confidence interval of 0.053 to 1.041 and P = 0.056. A landmark analysis at 12 weeks, performed to reduce immortal time bias, likewise found no significant association between seroconversion by that point and later events, with a hazard ratio of 0.822, a 95% confidence interval of 0.351 to 1.927 and P = 0.653. The authors noted that antibody production may sometimes follow immune-mediated thyroid injury rather than precede it, although the study was not designed to establish that mechanism.

Several limitations constrain the interpretation and generalizability of the findings. The cohort came from one hospital, was limited to advanced non-small-cell lung cancer and excluded people with known thyroid disorders, so it does not represent every population receiving immunotherapy. The number of baseline antibody-negative patients who both seroconverted and developed thyroid dysfunction was small, limiting statistical power. Thyrotropin-receptor antibodies were not measured, preventing complete differentiation of a small subset of Graves-like thyrotoxicosis, and the antibody and thyroid-stimulating hormone thresholds reflected the institution’s assay kits. The study focused on thyroid toxicity and collected neither tumor-survival nor treatment-efficacy outcomes. The 2.985 mIU/L thyroid-stimulating hormone threshold was derived internally and was explicitly considered exploratory rather than clinically actionable. Serial monitoring and integration of baseline thyroid measurements with dynamic assessment may ultimately assist risk stratification, but the study does not establish that either approach should independently guide clinical decisions.

Subject of Research: Prospective evaluation of baseline thyroid autoantibodies, TSH and post-treatment seroconversion as markers of immune-related thyroid dysfunction during advanced non-small-cell lung cancer immunotherapy

Article Title: Baseline markers and exploratory analysis of post-treatment thyroid autoantibody seroconversion for immune-related thyroid dysfunction in patients with advanced non-small-cell lung cancer treated with ICIs: a prospective single-center cohort study

Article References: Ming, H., Liang, Q., Hu, W., Chen, Y., & Yu, H. (2026). Baseline markers and exploratory analysis of post-treatment thyroid autoantibody seroconversion for immune-related thyroid dysfunction in patients with advanced non-small-cell lung cancer treated with ICIs: a prospective single-center cohort study. Frontiers in Oncology, 16, Article 1975574. https://doi.org/10.3389/fonc.2026.1975574

Image Credits: AI Generated

DOI: 10.3389/fonc.2026.1975574

Keywords: immune checkpoint inhibitors, thyroid dysfunction, thyroid autoantibodies, thyroid peroxidase antibodies, thyroglobulin antibodies, thyroid-stimulating hormone, antibody seroconversion, non-small-cell lung cancer, immune-related adverse events, prospective cohort, Baseline, markers